2 Aug
2021
2 Aug
'21
5:53 p.m.
On Aug 2, 2021, at 8:45 PM, Chuck Guzis via cctech
<cctech at classiccmp.org> wrote:
On 8/2/21 8:11 AM, James Liu via cctech wrote:
He did say "in the 1960s" so it may have been an early one without the high
quality optimizations that grew over time.
Thanks for feedback and offers to assist.
Happy to contirubte.
For some background, Tini Veltman developed
Schoonship in the 1960's
at CERN on the CDC 6600. My understanding is that he more or less
insisted on coding in assembly since he thought FORTRAN or other high
level languages would just get in the way and slow things down. The
code was maintained by Veltman and Strubbe well into the 1970's, but
its future was held back by being so closely tied to CDC hardware.
Which CDC FORTRAN? RUN, maybe--but FTN extended was pretty darned good
in optimizing and scheduling instructions. A lot of work went into that one. As a matter of fact, when we COMPASS scriveners came
up against a nasty
loop that we wanted to optimize for the 6600, one approach was to code
it in FORTRAN to see what the compiler would do with it and then work
from there. Some of the optimizations were quite startling,
particularly with the "UO" option selected.
If you've never written and hand-optimized 6600 code, it could be a
daunting task.
So I learned, having to do it on a 6400 -- which is quite a lot easier. But I learned a
lot from reading the OS source code, stuff like the analog of memcpy that used both the
boolean and shift units for transfer operations so it could do two of them one cycle
apart.
paul